One of the biggest requirements for document OCR is visual grounding, and frontier models (gemini, o...